1. Field of the Invention
The present invention relates to split-gate FLASH memory cells and, more particularly, to a split-gate FLASH memory cell with a tip in the middle of the floating gate.
2. Description of the Related Art
An electrically-erasable, programmable read-only-memory (EEPROM) cell is a semiconductor memory structure that stores one of two logic states even when power is removed from the structure. A conventional EEPROM cell is a two-transistor device that includes a select transistor and a floating-gate memory transistor.
A split-gate FLASH memory cell is a type of EEPROM cell that merges the select transistor and the floating-gate memory transistor into a single transistor. In a split-gate FLASH memory cell, a merged gate functions as both the gate of the select transistor and the control gate of the memory transistor.
FIG. 1A shows a plan view that illustrates a prior-art, split-gate FLASH memory cell 100. FIG. 1B shows a cross-sectional view taken along line 1B—1B of FIG. 1A. In the example shown in FIGS. 1A and 1B, memory cell 100 is formed in a p-substrate 110 that includes a pair of conventionally-formed, spaced-apart, shallow trench isolation regions STI.
Memory cell 100 includes spaced-apart n+ source and drain regions 112 and 114 that are formed in substrate 110, and a channel region 116 that is defined between source and drain regions 112 and 114. Channel region 116, in turn, includes a first channel region 116A, a second channel region 116B, and a third channel region 116C.
As further shown in FIGS. 1A and 1B, memory cell 100 includes a layer of gate oxide 120 that is formed on substrate 110 over channel region 116, and a floating gate 122 that is formed on gate oxide layer 120 over first channel region 116A. In addition, memory cell 100 also includes a layer of tunneling oxide 124 that is formed on gate oxide layer 120 over second channel region 116B, the side walls of floating gate 122, and the top surface of floating gate 122.
Memory cell 100 further includes a merged gate 126 that is formed on gate oxide layer 120 over third channel region 116C, and tunneling oxide layer 124 over floating gate 122. The portion of merged gate 126 that is formed over third channel region 116C functions like the gate of a select transistor of a conventional EEPROM cell. In addition, the portion of merged gate 126 that is formed over first channel region 116A functions like the control gate of a floating-gate memory transistor of a conventional EEPROM cell.
In operation, after memory cell 100 has been erased, floating gate 122 is left with a first charge. When memory cell 100 is then read, ground is applied to substrate 110 and source 112, a first positive voltage is applied to drain region 114, and a second positive voltage is applied to merged gate 126.
The first positive voltage sets up a drain-to-source electric field, while the second positive voltage sets up a merged gate-to-substrate electric field. The merged gate-to-substrate electric field in first channel region 116A is defined by several factors. These factors include the voltage differential, the separation distance, the dielectric constants of the materials, and the first charge stored on floating gate 122.
Under the influence of the merged gate-to-substrate electric field, electrons are attracted to and accumulate at the surface of channel region 116. The accumulated electrons, in turn, allow electrons to flow from source region 112 to drain region 114 under the influence of the drain-to-source electric field. The electron flow is detected and read to be, for example, a logic high.
After memory cell 100 has been programmed, a second charge is placed on floating gate 122. Memory cell 100 is programmed by injecting electrons into floating gate 122. As a result, the second charge is more negative than the first charge. As above, when memory cell 100 is read, ground is applied to substrate 110 and source 112, the first positive voltage is applied to drain region 114, and the second positive voltage is applied to control gate 126.
The first positive voltage sets up the same drain-to-source electric field. The second positive voltage, however, sets up a different merged gate-to-substrate electric field in first channel region 116A. In this case, the electric field is much weaker due to the negative value of the second charge stored on floating gate 122.
The weaker merged gate-to-substrate electric field, in turn, fails to attract enough electrons to the surface of first channel region 116A to allow electrons to flow from source region 112 to drain region 114. The lack of electron flow is detected and read to be, for example, a logic low.
Memory cell 100 is erased by applying a third positive voltage to merged gate 126, and ground or a negative voltage to source region 112. (The substrate and drain regions can be grounded or floated.) The floating gate 122 will also receive a part of third positive voltage in accordance with capacitive coupling of the floating gate to merged gate 126, source 112, and substrate 110.
The potential difference between the floating gate 122 and merged gate 126 sets up an electric field which has a magnitude that is sufficient to cause electrons on floating gate 122 to tunnel through tunneling oxide layer 124 to merged gate 126 via the well known Fowler-Nordheim process. The magnitude of the merged gate-to-floating gate electric field across tunneling oxide layer 124 is greatest at the covered edge of floating gate 122 where a tip 130 is present.
One problem with memory cell 100 is that when a large number of cells are fabricated, such as when wafers of memory arrays are formed, it is difficult to precisely replicate the steps that are used to form tip 130. In a conventional fabrication process, one side of a tip is formed by an anisotropic etch (which is difficult to align), while the other side is formed by the bird's beak portion of a local oxidation of silicon (LOCOS)-grown field oxide region (which is difficult to control).
The conventional process also leads to variations in the polysilicon grain structure, the radius of curvature of tip 130, the angle of tip 130, and the distance from tip 130 to the overlying tunneling oxide. As a result, tip 130 can vary significantly from lot-to-lot, thereby providing an undesirably wide performance range.
Thus, there is a need for a method of forming a split-gate memory cell that provides a more controllable approach to fabricating the memory cell and, therefore, a narrower performance range.